The desired workflow would run periodically (maybe once a day) and look something like this:
- Pull all new test files from the ClickHouse/ClickHouse repo
npm run generate:expected-ouputs to generate expected explain outputs for the new test files
- Delegate to an agent to make all of the tests pass
- Open (or update an existing) PR with the changes
We might need a way for the agent to determine if the new syntax is just unsupported (eg. contains kql or another parser limitation). We might also need to timebox the execution or otherwise ensure that the agent doesn't spin forever.
The desired workflow would run periodically (maybe once a day) and look something like this:
npm run generate:expected-ouputsto generate expected explain outputs for the new test filesWe might need a way for the agent to determine if the new syntax is just unsupported (eg. contains kql or another parser limitation). We might also need to timebox the execution or otherwise ensure that the agent doesn't spin forever.